Co-GAIL

We present a method for learning human-robot collaboration policy from human-human collaboration demonstrations. An effective robot assistant must learn to handle diverse human behaviors shown in the demonstrations and be robust when the humans adjust their strategies during online task execution. Our method co-optimizes a human policy and a robot policy in an interactive learning process: the human policy learns to generate diverse and plausible collaborative behaviors from demonstrations while the robot policy learns to assist by estimating the unobserved latent strategy of its human collaborator. Across a 2D strategy game, a human-robot handover task, and a multi-step collaborative manipulation task, our method outperforms the alternatives in both simulated evaluations and when executing the tasks with a real human operator in-the-loop.

We test our method and baselines in three task domains:

• A 2D low-dimensional collaborative game Fetch-Quest. (2D-Fetch-Quest)

• A high-dimensional human-robot handover. (HR-Handover)

• A multi-stage human-robot sequence manipulation task. (HR-SeqManip)

We aim to learn human-robot collaborative behaviors from human-human collaboration data. Our first step is to collect the human-human collaboration demonstrations.

For 2D-Fetch-Quest, the data is collected with two people using a pair of joysticks.

For HR-Handover and HR-SeqManip, we use the phone teleoperation platform RoboTurk. Both operators hold their phones to control the end-effectors of a humanoid and robot in the simulation to complete a collaboration task.

Here we show the real-human evaluation results in three task domains with different human operators. For the other experiment results (replay evaluation and interpolation), please refer to our paper and the supplementary video (top of this page).